A5039197772- Face recognition and analysis
- Face Recognition and Perception
- Evolutionary Psychology and Human Behavior
- Face and Expression Recognition
- Biometric Identification and Security
- Visual perception and processing mechanisms
- Visual Attention and Saliency Detection
- Video Surveillance and Tracking Methods
- Generative Adversarial Networks and Image Synthesis
- Neural dynamics and brain function
- Gait Recognition and Analysis
- Psychology of Moral and Emotional Judgment
- Deception detection and forensic psychology
- Neural and Behavioral Psychology Studies
- Facial Nerve Paralysis Treatment and Research
- Aesthetic Perception and Analysis
- Noise Effects and Management
- Hearing Loss and Rehabilitation
- Emotion and Mood Recognition
- Memory Processes and Influences
- Consumer Perception and Purchasing Behavior
- Color Science and Applications
- Neural Networks and Applications
- 3D Shape Modeling and Analysis
- Forensic Anthropology and Bioarchaeology Studies
The University of Texas at Dallas
2015-2024
ORCID
2024
University of California, Los Angeles
2014
National Institute of Standards and Technology
1999-2011
Max Planck Society
1995-2010
Max Planck Institute for Biological Cybernetics
1995-2010
The University of Texas at Austin
2002-2005
In-Q-Tel
2002
Brown University
1986-1989
John Brown University
1987
This paper describes the large-scale experimental results from Face Recognition Vendor Test (FRVT) 2006 and Iris Challenge Evaluation (ICE) 2006. The FRVT looked at recognition high-resolution still frontal face images 3D images, measured performance for taken under controlled uncontrolled illumination. ICE evaluation reported verification both left right irises. in intentionally represent a broader range of quality than sensor would normally acquire. includes that did not pass control...
Significance This study measures face identification accuracy for an international group of professional forensic facial examiners working under circumstances that apply in real world casework. Examiners and other human “specialists,” including forensically trained reviewers untrained superrecognizers, were more accurate than the control groups on a challenging test identification. Therefore, specialists are best available solution to problem We present data comparing state-of-the-art...
Forensic facial identification examiners are required to match the identity of faces in images that vary substantially, owing changes viewing conditions and a person's appearance. These identifications affect course outcome criminal investigations convictions. Despite calls for research on sources human error forensic examination, existing scientific knowledge face matching accuracy is based, almost exclusively, people without formal training. Here, we administered three challenging tests...
Previous generations of face recognition algorithms differ in accuracy for images different races (race bias). Here, we present the possible underlying factors (data-driven and scenario modeling) methodological considerations assessing race bias algorithms. We discuss data-driven (e.g., image quality, population statistics, algorithm architecture), modeling that consider role "user" threshold decisions demographic constraints). To illustrate how these issues apply, data from four (a...
Abstract Object and face representations in ventral temporal (VT) cortex were investigated by combining object confusability data from a computational model of classification with neural response functional neuroimaging experiment. A pattern-based algorithm learned to categorize individual brain maps according the category being viewed subject. An identical classify an image-based, view-dependent representation stimuli. High correlations found between categories activity maps. This occurred...
Faces can be represented efficiently as a weighted linear combination of the eigenvectors covariance matrix face images. It has also been shown [ J. Opt. Soc. Am.4, 519– 524 ( 1987)] that identifiable faces made by using only subset eigenvectors, i.e., those with largest eigenvalues. This low-dimensional representation is optimal in it minimizes squared error between image and original image. The present study demonstrates that, whereas this for identifying physical categories face, like...
There has been significant progress in improving the performance of computer-based face recognition algorithms over last decade. Although have tested and compared extensively with each other, there remarkably little work comparing accuracy systems humans. We seven state-of-the-art humans on a facematching task. Humans determined whether pairs images, taken under different illumination conditions, were pictures same person or people. Three surpassed human matching prescreened to be...
We describe a database of static images and video clips human faces people that is useful for testing algorithms face person recognition, head/eye tracking, computer graphics modeling natural motions. For each there are nine "facial mug shots" series streams. The videos include "moving facial shot," speech clip, one or more dynamic expression clips, two gait videos, conversation taken at moderate distance from the camera. Complete data sets available 284 subjects duplicate sets, subsequent...
Psychological research indicates that humans recognize faces of their own race more accurately than other races. This “other-race effect” occurs for algorithms tested in a recent international competition state-of-the-art face recognition algorithms. We report results Western algorithm made by fusing eight from countries and an East Asian five countries. At the low false accept rates required most security applications, recognized Caucasian faces. Next, using test spanned all alarm rates, we...
The Good, the Bad, & Ugly Face Challenge Problem was created to encourage development of algorithms that are robust recognition across changes occur in still frontal faces. consists three partitions. Good partition contains pairs images considered easy recognize. On partition, base verification rate (VR) is 0.98 at a false accept (FAR) 0.001. Bad average difficulty For VR 0.80 FAR difficult recognize, with 0.15 performance from fusing output top performers FRVT 2006. design controls for pose...